Mental imagery is the mental re-creation of perceptual experiences, events and scenarios, and motor acts. In our previous study, we assessed whether motor imagery (MI) training combined with functional magnetic resonance imaging-based neurofeedback could improve the motor function of nondemented subjects with mild Parkinson's disease (PD) (N=22). We used visual imagery (VI) (e.g., of scenes or events, but not of self-movements) training without neurofeedback for the control group (N=22). Notably, both groups showed significant and comparable improvement in motor function after four weeks of daily imagery practice. In this study, we further examined the neural correlates of the motor enhancement as a result of the VI training by analyzing the self-reported VI content during daily practice and relating its quality to the functional connectivity characteristics of the same subjects. We demonstrated that the VI practice encompassed multisensory, spatial, affective, and executive processes all of which are also important for motor function in real life. Subjects with worse global disease severity also showed poorer quality of the VI content. Finally, the quality of the VI content showed significant positive correlations with the functional connectivity changes during the VI tasks in brain areas supporting visuospatial and sensorimotor processes. Our findings suggest that mental imagery training combining VI and MI may enhance motor function in patients with mild PD, and more broadly, underline the importance of incorporating self-reports of thoughts and experiences in neuroimaging studies that examine the brain mechanisms of complex cognitive processes especially in neuropsychiatric patient populations.